Invisible fault identifying in loess area is a difficult problem in active fault study in northern China. Detailed stratigraphic division of loess area by the naked eye is very difficult due to the insignificant difference of the granularities and the colors,which would affect the identification of the obscured fault and paleo - seismic event. Spectral technique has been used for magnetic susceptibility estimation. Magnetic susceptibility (MS) has been considered to be a measure of the degree of pedogenic activity and excellent proxies for terrestrial climatic fluctuations. In this study,multiple linear regression was used to build MS estimation models based on the spectral features. A model was built and was applied to hyperspectral image. Test of datasets indicates that this model is very successful. The applying of this model to hyperspectral image shows that the intensity distribution of MS could be used for stratigraphic division.